lodrick-the-lafted/Hermes-100K

---
language:
- eng
pretty_name: Hermes-100K
tags:
- distillation
- synthetic data
- gpt
task_categories:
- text-generation
---

It's 100K rows sampled from teknium/openhermes (not the newer 2.5).

Filtered some GPTisms I dislike out, and removed rows with short output as well to bias towards longer answers.

bad_phrases = ["couldn't help but", "can't resist", "random", "unethical", "I'm sorry, but", "I'm sorry but", "as an AI", "as a Language Model", "AI Language Model", "language model", "However, it is important to", "However, it's important", "ethical guidelines", "just an AI", "within my programming", "illegal", "cannot provide"]

原始信息汇总

数据集概述

基本信息

• 语言: 英语
• 名称: Hermes-100K
• 标签:
  • 蒸馏 (distillation)
  • 合成数据 (synthetic data)
  • GPT
• 任务类别: 文本生成 (text-generation)

数据来源

• 数据集是从 teknium/openhermes（非更新的2.5版本）中采样的100,000行数据。

数据处理

• 过滤了一些不受欢迎的表达方式。
• 移除了输出较短的行，以偏向于较长的回答。

过滤词组

• couldnt help but
• cant resist
• random
• unethical
• Im sorry, but
• Im sorry but
• as an AI
• as a Language Model
• AI Language Model
• language model
• However, it is important to
• However, its important
• ethical guidelines
• just an AI
• within my programming
• illegal
• cannot provide